| Benchmark | Kimi K2 (Reasoning) | o4-mini |
|---|---|---|
| AA Agentic Index | 47.9 | 36.1 |
| AA Intelligence | 33.5 | 26.1 |
| AA-LCR | 70.3 | 60 |
| AA-Omniscience | -21.4 | -35.7 |
| aider_polyglot | 59.1 | 72 |
| AIME 2025 | 94.5 | 92.7 |
| Artificial Analysis Coding Index | 34.8 | 25.6 |
| browsecomp | 60.2 | 51.5 |
| critpt | 2.6 | 0.6 |
| frontiermath_tier_4 | 0 | 6.3 |
| gdpval | 24.5 | 25.4 |
| GPQA Diamond | 84.5 | 81.4 |
| HLE | 23.9 | 16.5 |
| IFBench | 68.1 | 68.7 |
| LiveCodeBench | 79.2 | 84.5 |
| multichallenge | 66.4 | 44.9 |
| OmniScience Accuracy | 30.9 | 24.8 |
| OmniScience Non-Hallucination | 25.8 | 19.5 |
| scicode | 44.8 | 46.5 |
| simplebench | 39.6 | 38.7 |
| swe_bench_bash | 63.4 | 45 |
| SWE-bench Verified | 71.3 | 68.1 |
| Terminal-Bench Hard | 31.1 | 15.2 |
| vectara_answer_rate | 98.6 | 99.2 |
| vectara_avg_summary_length | 59.2 | 130.9 |
| vectara_factual_consistency | 82.1 | 81.4 |
| vectara_hallucination_rate ↓ | 17.9 | 18.6 |
| τ²-Bench Telecom (AA run) | 93 | 55.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.